2 * mm/rmap.c - physical to virtual reverse mappings
4 * Copyright 2001, Rik van Riel <riel@conectiva.com.br>
5 * Released under the General Public License (GPL).
7 * Simple, low overhead reverse mapping scheme.
8 * Please try to keep this thing as modular as possible.
10 * Provides methods for unmapping each kind of mapped page:
11 * the anon methods track anonymous pages, and
12 * the file methods track pages belonging to an inode.
14 * Original design by Rik van Riel <riel@conectiva.com.br> 2001
15 * File methods by Dave McCracken <dmccr@us.ibm.com> 2003, 2004
16 * Anonymous methods by Andrea Arcangeli <andrea@suse.de> 2004
17 * Contributions by Hugh Dickins 2003, 2004
21 * Lock ordering in mm:
23 * inode->i_mutex (while writing or truncating, not reading or faulting)
25 * page->flags PG_locked (lock_page)
26 * hugetlbfs_i_mmap_rwsem_key (in huge_pmd_share)
27 * mapping->i_mmap_rwsem
29 * mm->page_table_lock or pte_lock
30 * zone_lru_lock (in mark_page_accessed, isolate_lru_page)
31 * swap_lock (in swap_duplicate, swap_info_get)
32 * mmlist_lock (in mmput, drain_mmlist and others)
33 * mapping->private_lock (in __set_page_dirty_buffers)
34 * mem_cgroup_{begin,end}_page_stat (memcg->move_lock)
35 * mapping->tree_lock (widely used)
36 * inode->i_lock (in set_page_dirty's __mark_inode_dirty)
37 * bdi.wb->list_lock (in set_page_dirty's __mark_inode_dirty)
38 * sb_lock (within inode_lock in fs/fs-writeback.c)
39 * mapping->tree_lock (widely used, in set_page_dirty,
40 * in arch-dependent flush_dcache_mmap_lock,
41 * within bdi.wb->list_lock in __sync_single_inode)
43 * anon_vma->rwsem,mapping->i_mutex (memory_failure, collect_procs_anon)
49 #include <linux/sched/mm.h>
50 #include <linux/sched/task.h>
51 #include <linux/pagemap.h>
52 #include <linux/swap.h>
53 #include <linux/swapops.h>
54 #include <linux/slab.h>
55 #include <linux/init.h>
56 #include <linux/ksm.h>
57 #include <linux/rmap.h>
58 #include <linux/rcupdate.h>
59 #include <linux/export.h>
60 #include <linux/memcontrol.h>
61 #include <linux/mmu_notifier.h>
62 #include <linux/migrate.h>
63 #include <linux/hugetlb.h>
64 #include <linux/backing-dev.h>
65 #include <linux/page_idle.h>
67 #include <asm/tlbflush.h>
69 #include <trace/events/tlb.h>
73 static struct kmem_cache
*anon_vma_cachep
;
74 static struct kmem_cache
*anon_vma_chain_cachep
;
76 static inline struct anon_vma
*anon_vma_alloc(void)
78 struct anon_vma
*anon_vma
;
80 anon_vma
= kmem_cache_alloc(anon_vma_cachep
, GFP_KERNEL
);
82 atomic_set(&anon_vma
->refcount
, 1);
83 anon_vma
->degree
= 1; /* Reference for first vma */
84 anon_vma
->parent
= anon_vma
;
86 * Initialise the anon_vma root to point to itself. If called
87 * from fork, the root will be reset to the parents anon_vma.
89 anon_vma
->root
= anon_vma
;
95 static inline void anon_vma_free(struct anon_vma
*anon_vma
)
97 VM_BUG_ON(atomic_read(&anon_vma
->refcount
));
100 * Synchronize against page_lock_anon_vma_read() such that
101 * we can safely hold the lock without the anon_vma getting
104 * Relies on the full mb implied by the atomic_dec_and_test() from
105 * put_anon_vma() against the acquire barrier implied by
106 * down_read_trylock() from page_lock_anon_vma_read(). This orders:
108 * page_lock_anon_vma_read() VS put_anon_vma()
109 * down_read_trylock() atomic_dec_and_test()
111 * atomic_read() rwsem_is_locked()
113 * LOCK should suffice since the actual taking of the lock must
114 * happen _before_ what follows.
117 if (rwsem_is_locked(&anon_vma
->root
->rwsem
)) {
118 anon_vma_lock_write(anon_vma
);
119 anon_vma_unlock_write(anon_vma
);
122 kmem_cache_free(anon_vma_cachep
, anon_vma
);
125 static inline struct anon_vma_chain
*anon_vma_chain_alloc(gfp_t gfp
)
127 return kmem_cache_alloc(anon_vma_chain_cachep
, gfp
);
130 static void anon_vma_chain_free(struct anon_vma_chain
*anon_vma_chain
)
132 kmem_cache_free(anon_vma_chain_cachep
, anon_vma_chain
);
135 static void anon_vma_chain_link(struct vm_area_struct
*vma
,
136 struct anon_vma_chain
*avc
,
137 struct anon_vma
*anon_vma
)
140 avc
->anon_vma
= anon_vma
;
141 list_add(&avc
->same_vma
, &vma
->anon_vma_chain
);
142 anon_vma_interval_tree_insert(avc
, &anon_vma
->rb_root
);
146 * __anon_vma_prepare - attach an anon_vma to a memory region
147 * @vma: the memory region in question
149 * This makes sure the memory mapping described by 'vma' has
150 * an 'anon_vma' attached to it, so that we can associate the
151 * anonymous pages mapped into it with that anon_vma.
153 * The common case will be that we already have one, which
154 * is handled inline by anon_vma_prepare(). But if
155 * not we either need to find an adjacent mapping that we
156 * can re-use the anon_vma from (very common when the only
157 * reason for splitting a vma has been mprotect()), or we
158 * allocate a new one.
160 * Anon-vma allocations are very subtle, because we may have
161 * optimistically looked up an anon_vma in page_lock_anon_vma_read()
162 * and that may actually touch the spinlock even in the newly
163 * allocated vma (it depends on RCU to make sure that the
164 * anon_vma isn't actually destroyed).
166 * As a result, we need to do proper anon_vma locking even
167 * for the new allocation. At the same time, we do not want
168 * to do any locking for the common case of already having
171 * This must be called with the mmap_sem held for reading.
173 int __anon_vma_prepare(struct vm_area_struct
*vma
)
175 struct mm_struct
*mm
= vma
->vm_mm
;
176 struct anon_vma
*anon_vma
, *allocated
;
177 struct anon_vma_chain
*avc
;
181 avc
= anon_vma_chain_alloc(GFP_KERNEL
);
185 anon_vma
= find_mergeable_anon_vma(vma
);
188 anon_vma
= anon_vma_alloc();
189 if (unlikely(!anon_vma
))
190 goto out_enomem_free_avc
;
191 allocated
= anon_vma
;
194 anon_vma_lock_write(anon_vma
);
195 /* page_table_lock to protect against threads */
196 spin_lock(&mm
->page_table_lock
);
197 if (likely(!vma
->anon_vma
)) {
198 vma
->anon_vma
= anon_vma
;
199 anon_vma_chain_link(vma
, avc
, anon_vma
);
200 /* vma reference or self-parent link for new root */
205 spin_unlock(&mm
->page_table_lock
);
206 anon_vma_unlock_write(anon_vma
);
208 if (unlikely(allocated
))
209 put_anon_vma(allocated
);
211 anon_vma_chain_free(avc
);
216 anon_vma_chain_free(avc
);
222 * This is a useful helper function for locking the anon_vma root as
223 * we traverse the vma->anon_vma_chain, looping over anon_vma's that
226 * Such anon_vma's should have the same root, so you'd expect to see
227 * just a single mutex_lock for the whole traversal.
229 static inline struct anon_vma
*lock_anon_vma_root(struct anon_vma
*root
, struct anon_vma
*anon_vma
)
231 struct anon_vma
*new_root
= anon_vma
->root
;
232 if (new_root
!= root
) {
233 if (WARN_ON_ONCE(root
))
234 up_write(&root
->rwsem
);
236 down_write(&root
->rwsem
);
241 static inline void unlock_anon_vma_root(struct anon_vma
*root
)
244 up_write(&root
->rwsem
);
248 * Attach the anon_vmas from src to dst.
249 * Returns 0 on success, -ENOMEM on failure.
251 * If dst->anon_vma is NULL this function tries to find and reuse existing
252 * anon_vma which has no vmas and only one child anon_vma. This prevents
253 * degradation of anon_vma hierarchy to endless linear chain in case of
254 * constantly forking task. On the other hand, an anon_vma with more than one
255 * child isn't reused even if there was no alive vma, thus rmap walker has a
256 * good chance of avoiding scanning the whole hierarchy when it searches where
259 int anon_vma_clone(struct vm_area_struct
*dst
, struct vm_area_struct
*src
)
261 struct anon_vma_chain
*avc
, *pavc
;
262 struct anon_vma
*root
= NULL
;
264 list_for_each_entry_reverse(pavc
, &src
->anon_vma_chain
, same_vma
) {
265 struct anon_vma
*anon_vma
;
267 avc
= anon_vma_chain_alloc(GFP_NOWAIT
| __GFP_NOWARN
);
268 if (unlikely(!avc
)) {
269 unlock_anon_vma_root(root
);
271 avc
= anon_vma_chain_alloc(GFP_KERNEL
);
275 anon_vma
= pavc
->anon_vma
;
276 root
= lock_anon_vma_root(root
, anon_vma
);
277 anon_vma_chain_link(dst
, avc
, anon_vma
);
280 * Reuse existing anon_vma if its degree lower than two,
281 * that means it has no vma and only one anon_vma child.
283 * Do not chose parent anon_vma, otherwise first child
284 * will always reuse it. Root anon_vma is never reused:
285 * it has self-parent reference and at least one child.
287 if (!dst
->anon_vma
&& anon_vma
!= src
->anon_vma
&&
288 anon_vma
->degree
< 2)
289 dst
->anon_vma
= anon_vma
;
292 dst
->anon_vma
->degree
++;
293 unlock_anon_vma_root(root
);
298 * dst->anon_vma is dropped here otherwise its degree can be incorrectly
299 * decremented in unlink_anon_vmas().
300 * We can safely do this because callers of anon_vma_clone() don't care
301 * about dst->anon_vma if anon_vma_clone() failed.
303 dst
->anon_vma
= NULL
;
304 unlink_anon_vmas(dst
);
309 * Attach vma to its own anon_vma, as well as to the anon_vmas that
310 * the corresponding VMA in the parent process is attached to.
311 * Returns 0 on success, non-zero on failure.
313 int anon_vma_fork(struct vm_area_struct
*vma
, struct vm_area_struct
*pvma
)
315 struct anon_vma_chain
*avc
;
316 struct anon_vma
*anon_vma
;
319 /* Don't bother if the parent process has no anon_vma here. */
323 /* Drop inherited anon_vma, we'll reuse existing or allocate new. */
324 vma
->anon_vma
= NULL
;
327 * First, attach the new VMA to the parent VMA's anon_vmas,
328 * so rmap can find non-COWed pages in child processes.
330 error
= anon_vma_clone(vma
, pvma
);
334 /* An existing anon_vma has been reused, all done then. */
338 /* Then add our own anon_vma. */
339 anon_vma
= anon_vma_alloc();
342 avc
= anon_vma_chain_alloc(GFP_KERNEL
);
344 goto out_error_free_anon_vma
;
347 * The root anon_vma's spinlock is the lock actually used when we
348 * lock any of the anon_vmas in this anon_vma tree.
350 anon_vma
->root
= pvma
->anon_vma
->root
;
351 anon_vma
->parent
= pvma
->anon_vma
;
353 * With refcounts, an anon_vma can stay around longer than the
354 * process it belongs to. The root anon_vma needs to be pinned until
355 * this anon_vma is freed, because the lock lives in the root.
357 get_anon_vma(anon_vma
->root
);
358 /* Mark this anon_vma as the one where our new (COWed) pages go. */
359 vma
->anon_vma
= anon_vma
;
360 anon_vma_lock_write(anon_vma
);
361 anon_vma_chain_link(vma
, avc
, anon_vma
);
362 anon_vma
->parent
->degree
++;
363 anon_vma_unlock_write(anon_vma
);
367 out_error_free_anon_vma
:
368 put_anon_vma(anon_vma
);
370 unlink_anon_vmas(vma
);
374 void unlink_anon_vmas(struct vm_area_struct
*vma
)
376 struct anon_vma_chain
*avc
, *next
;
377 struct anon_vma
*root
= NULL
;
380 * Unlink each anon_vma chained to the VMA. This list is ordered
381 * from newest to oldest, ensuring the root anon_vma gets freed last.
383 list_for_each_entry_safe(avc
, next
, &vma
->anon_vma_chain
, same_vma
) {
384 struct anon_vma
*anon_vma
= avc
->anon_vma
;
386 root
= lock_anon_vma_root(root
, anon_vma
);
387 anon_vma_interval_tree_remove(avc
, &anon_vma
->rb_root
);
390 * Leave empty anon_vmas on the list - we'll need
391 * to free them outside the lock.
393 if (RB_EMPTY_ROOT(&anon_vma
->rb_root
)) {
394 anon_vma
->parent
->degree
--;
398 list_del(&avc
->same_vma
);
399 anon_vma_chain_free(avc
);
402 vma
->anon_vma
->degree
--;
403 unlock_anon_vma_root(root
);
406 * Iterate the list once more, it now only contains empty and unlinked
407 * anon_vmas, destroy them. Could not do before due to __put_anon_vma()
408 * needing to write-acquire the anon_vma->root->rwsem.
410 list_for_each_entry_safe(avc
, next
, &vma
->anon_vma_chain
, same_vma
) {
411 struct anon_vma
*anon_vma
= avc
->anon_vma
;
413 VM_WARN_ON(anon_vma
->degree
);
414 put_anon_vma(anon_vma
);
416 list_del(&avc
->same_vma
);
417 anon_vma_chain_free(avc
);
421 static void anon_vma_ctor(void *data
)
423 struct anon_vma
*anon_vma
= data
;
425 init_rwsem(&anon_vma
->rwsem
);
426 atomic_set(&anon_vma
->refcount
, 0);
427 anon_vma
->rb_root
= RB_ROOT
;
430 void __init
anon_vma_init(void)
432 anon_vma_cachep
= kmem_cache_create("anon_vma", sizeof(struct anon_vma
),
433 0, SLAB_TYPESAFE_BY_RCU
|SLAB_PANIC
|SLAB_ACCOUNT
,
435 anon_vma_chain_cachep
= KMEM_CACHE(anon_vma_chain
,
436 SLAB_PANIC
|SLAB_ACCOUNT
);
440 * Getting a lock on a stable anon_vma from a page off the LRU is tricky!
442 * Since there is no serialization what so ever against page_remove_rmap()
443 * the best this function can do is return a locked anon_vma that might
444 * have been relevant to this page.
446 * The page might have been remapped to a different anon_vma or the anon_vma
447 * returned may already be freed (and even reused).
449 * In case it was remapped to a different anon_vma, the new anon_vma will be a
450 * child of the old anon_vma, and the anon_vma lifetime rules will therefore
451 * ensure that any anon_vma obtained from the page will still be valid for as
452 * long as we observe page_mapped() [ hence all those page_mapped() tests ].
454 * All users of this function must be very careful when walking the anon_vma
455 * chain and verify that the page in question is indeed mapped in it
456 * [ something equivalent to page_mapped_in_vma() ].
458 * Since anon_vma's slab is DESTROY_BY_RCU and we know from page_remove_rmap()
459 * that the anon_vma pointer from page->mapping is valid if there is a
460 * mapcount, we can dereference the anon_vma after observing those.
462 struct anon_vma
*page_get_anon_vma(struct page
*page
)
464 struct anon_vma
*anon_vma
= NULL
;
465 unsigned long anon_mapping
;
468 anon_mapping
= (unsigned long)READ_ONCE(page
->mapping
);
469 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
471 if (!page_mapped(page
))
474 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
475 if (!atomic_inc_not_zero(&anon_vma
->refcount
)) {
481 * If this page is still mapped, then its anon_vma cannot have been
482 * freed. But if it has been unmapped, we have no security against the
483 * anon_vma structure being freed and reused (for another anon_vma:
484 * SLAB_TYPESAFE_BY_RCU guarantees that - so the atomic_inc_not_zero()
485 * above cannot corrupt).
487 if (!page_mapped(page
)) {
489 put_anon_vma(anon_vma
);
499 * Similar to page_get_anon_vma() except it locks the anon_vma.
501 * Its a little more complex as it tries to keep the fast path to a single
502 * atomic op -- the trylock. If we fail the trylock, we fall back to getting a
503 * reference like with page_get_anon_vma() and then block on the mutex.
505 struct anon_vma
*page_lock_anon_vma_read(struct page
*page
)
507 struct anon_vma
*anon_vma
= NULL
;
508 struct anon_vma
*root_anon_vma
;
509 unsigned long anon_mapping
;
512 anon_mapping
= (unsigned long)READ_ONCE(page
->mapping
);
513 if ((anon_mapping
& PAGE_MAPPING_FLAGS
) != PAGE_MAPPING_ANON
)
515 if (!page_mapped(page
))
518 anon_vma
= (struct anon_vma
*) (anon_mapping
- PAGE_MAPPING_ANON
);
519 root_anon_vma
= READ_ONCE(anon_vma
->root
);
520 if (down_read_trylock(&root_anon_vma
->rwsem
)) {
522 * If the page is still mapped, then this anon_vma is still
523 * its anon_vma, and holding the mutex ensures that it will
524 * not go away, see anon_vma_free().
526 if (!page_mapped(page
)) {
527 up_read(&root_anon_vma
->rwsem
);
533 /* trylock failed, we got to sleep */
534 if (!atomic_inc_not_zero(&anon_vma
->refcount
)) {
539 if (!page_mapped(page
)) {
541 put_anon_vma(anon_vma
);
545 /* we pinned the anon_vma, its safe to sleep */
547 anon_vma_lock_read(anon_vma
);
549 if (atomic_dec_and_test(&anon_vma
->refcount
)) {
551 * Oops, we held the last refcount, release the lock
552 * and bail -- can't simply use put_anon_vma() because
553 * we'll deadlock on the anon_vma_lock_write() recursion.
555 anon_vma_unlock_read(anon_vma
);
556 __put_anon_vma(anon_vma
);
567 void page_unlock_anon_vma_read(struct anon_vma
*anon_vma
)
569 anon_vma_unlock_read(anon_vma
);
572 #ifdef CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH
574 * Flush TLB entries for recently unmapped pages from remote CPUs. It is
575 * important if a PTE was dirty when it was unmapped that it's flushed
576 * before any IO is initiated on the page to prevent lost writes. Similarly,
577 * it must be flushed before freeing to prevent data leakage.
579 void try_to_unmap_flush(void)
581 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
583 if (!tlb_ubc
->flush_required
)
586 arch_tlbbatch_flush(&tlb_ubc
->arch
);
587 tlb_ubc
->flush_required
= false;
588 tlb_ubc
->writable
= false;
591 /* Flush iff there are potentially writable TLB entries that can race with IO */
592 void try_to_unmap_flush_dirty(void)
594 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
596 if (tlb_ubc
->writable
)
597 try_to_unmap_flush();
600 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
, bool writable
)
602 struct tlbflush_unmap_batch
*tlb_ubc
= ¤t
->tlb_ubc
;
604 arch_tlbbatch_add_mm(&tlb_ubc
->arch
, mm
);
605 tlb_ubc
->flush_required
= true;
608 * Ensure compiler does not re-order the setting of tlb_flush_batched
609 * before the PTE is cleared.
612 mm
->tlb_flush_batched
= true;
615 * If the PTE was dirty then it's best to assume it's writable. The
616 * caller must use try_to_unmap_flush_dirty() or try_to_unmap_flush()
617 * before the page is queued for IO.
620 tlb_ubc
->writable
= true;
624 * Returns true if the TLB flush should be deferred to the end of a batch of
625 * unmap operations to reduce IPIs.
627 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
629 bool should_defer
= false;
631 if (!(flags
& TTU_BATCH_FLUSH
))
634 /* If remote CPUs need to be flushed then defer batch the flush */
635 if (cpumask_any_but(mm_cpumask(mm
), get_cpu()) < nr_cpu_ids
)
643 * Reclaim unmaps pages under the PTL but do not flush the TLB prior to
644 * releasing the PTL if TLB flushes are batched. It's possible for a parallel
645 * operation such as mprotect or munmap to race between reclaim unmapping
646 * the page and flushing the page. If this race occurs, it potentially allows
647 * access to data via a stale TLB entry. Tracking all mm's that have TLB
648 * batching in flight would be expensive during reclaim so instead track
649 * whether TLB batching occurred in the past and if so then do a flush here
650 * if required. This will cost one additional flush per reclaim cycle paid
651 * by the first operation at risk such as mprotect and mumap.
653 * This must be called under the PTL so that an access to tlb_flush_batched
654 * that is potentially a "reclaim vs mprotect/munmap/etc" race will synchronise
657 void flush_tlb_batched_pending(struct mm_struct
*mm
)
659 if (mm
->tlb_flush_batched
) {
663 * Do not allow the compiler to re-order the clearing of
664 * tlb_flush_batched before the tlb is flushed.
667 mm
->tlb_flush_batched
= false;
671 static void set_tlb_ubc_flush_pending(struct mm_struct
*mm
, bool writable
)
675 static bool should_defer_flush(struct mm_struct
*mm
, enum ttu_flags flags
)
679 #endif /* CONFIG_ARCH_WANT_BATCHED_UNMAP_TLB_FLUSH */
682 * At what user virtual address is page expected in vma?
683 * Caller should check the page is actually part of the vma.
685 unsigned long page_address_in_vma(struct page
*page
, struct vm_area_struct
*vma
)
687 unsigned long address
;
688 if (PageAnon(page
)) {
689 struct anon_vma
*page__anon_vma
= page_anon_vma(page
);
691 * Note: swapoff's unuse_vma() is more efficient with this
692 * check, and needs it to match anon_vma when KSM is active.
694 if (!vma
->anon_vma
|| !page__anon_vma
||
695 vma
->anon_vma
->root
!= page__anon_vma
->root
)
697 } else if (page
->mapping
) {
698 if (!vma
->vm_file
|| vma
->vm_file
->f_mapping
!= page
->mapping
)
702 address
= __vma_address(page
, vma
);
703 if (unlikely(address
< vma
->vm_start
|| address
>= vma
->vm_end
))
708 pmd_t
*mm_find_pmd(struct mm_struct
*mm
, unsigned long address
)
716 pgd
= pgd_offset(mm
, address
);
717 if (!pgd_present(*pgd
))
720 p4d
= p4d_offset(pgd
, address
);
721 if (!p4d_present(*p4d
))
724 pud
= pud_offset(p4d
, address
);
725 if (!pud_present(*pud
))
728 pmd
= pmd_offset(pud
, address
);
730 * Some THP functions use the sequence pmdp_huge_clear_flush(), set_pmd_at()
731 * without holding anon_vma lock for write. So when looking for a
732 * genuine pmde (in which to find pte), test present and !THP together.
736 if (!pmd_present(pmde
) || pmd_trans_huge(pmde
))
742 struct page_referenced_arg
{
745 unsigned long vm_flags
;
746 struct mem_cgroup
*memcg
;
749 * arg: page_referenced_arg will be passed
751 static bool page_referenced_one(struct page
*page
, struct vm_area_struct
*vma
,
752 unsigned long address
, void *arg
)
754 struct page_referenced_arg
*pra
= arg
;
755 struct page_vma_mapped_walk pvmw
= {
762 while (page_vma_mapped_walk(&pvmw
)) {
763 address
= pvmw
.address
;
765 if (vma
->vm_flags
& VM_LOCKED
) {
766 page_vma_mapped_walk_done(&pvmw
);
767 pra
->vm_flags
|= VM_LOCKED
;
768 return false; /* To break the loop */
772 if (ptep_clear_flush_young_notify(vma
, address
,
775 * Don't treat a reference through
776 * a sequentially read mapping as such.
777 * If the page has been used in another mapping,
778 * we will catch it; if this other mapping is
779 * already gone, the unmap path will have set
780 * PG_referenced or activated the page.
782 if (likely(!(vma
->vm_flags
& VM_SEQ_READ
)))
785 } else if (IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
)) {
786 if (pmdp_clear_flush_young_notify(vma
, address
,
790 /* unexpected pmd-mapped page? */
798 clear_page_idle(page
);
799 if (test_and_clear_page_young(page
))
804 pra
->vm_flags
|= vma
->vm_flags
;
808 return false; /* To break the loop */
813 static bool invalid_page_referenced_vma(struct vm_area_struct
*vma
, void *arg
)
815 struct page_referenced_arg
*pra
= arg
;
816 struct mem_cgroup
*memcg
= pra
->memcg
;
818 if (!mm_match_cgroup(vma
->vm_mm
, memcg
))
825 * page_referenced - test if the page was referenced
826 * @page: the page to test
827 * @is_locked: caller holds lock on the page
828 * @memcg: target memory cgroup
829 * @vm_flags: collect encountered vma->vm_flags who actually referenced the page
831 * Quick test_and_clear_referenced for all mappings to a page,
832 * returns the number of ptes which referenced the page.
834 int page_referenced(struct page
*page
,
836 struct mem_cgroup
*memcg
,
837 unsigned long *vm_flags
)
840 struct page_referenced_arg pra
= {
841 .mapcount
= total_mapcount(page
),
844 struct rmap_walk_control rwc
= {
845 .rmap_one
= page_referenced_one
,
847 .anon_lock
= page_lock_anon_vma_read
,
851 if (!page_mapped(page
))
854 if (!page_rmapping(page
))
857 if (!is_locked
&& (!PageAnon(page
) || PageKsm(page
))) {
858 we_locked
= trylock_page(page
);
864 * If we are reclaiming on behalf of a cgroup, skip
865 * counting on behalf of references from different
869 rwc
.invalid_vma
= invalid_page_referenced_vma
;
872 rmap_walk(page
, &rwc
);
873 *vm_flags
= pra
.vm_flags
;
878 return pra
.referenced
;
881 static bool page_mkclean_one(struct page
*page
, struct vm_area_struct
*vma
,
882 unsigned long address
, void *arg
)
884 struct page_vma_mapped_walk pvmw
= {
891 bool invalidation_needed
= false;
893 while (page_vma_mapped_walk(&pvmw
)) {
897 pte_t
*pte
= pvmw
.pte
;
899 if (!pte_dirty(*pte
) && !pte_write(*pte
))
902 flush_cache_page(vma
, pvmw
.address
, pte_pfn(*pte
));
903 entry
= ptep_clear_flush(vma
, pvmw
.address
, pte
);
904 entry
= pte_wrprotect(entry
);
905 entry
= pte_mkclean(entry
);
906 set_pte_at(vma
->vm_mm
, pvmw
.address
, pte
, entry
);
909 #ifdef CONFIG_TRANSPARENT_HUGE_PAGECACHE
910 pmd_t
*pmd
= pvmw
.pmd
;
913 if (!pmd_dirty(*pmd
) && !pmd_write(*pmd
))
916 flush_cache_page(vma
, pvmw
.address
, page_to_pfn(page
));
917 entry
= pmdp_huge_clear_flush(vma
, pvmw
.address
, pmd
);
918 entry
= pmd_wrprotect(entry
);
919 entry
= pmd_mkclean(entry
);
920 set_pmd_at(vma
->vm_mm
, pvmw
.address
, pmd
, entry
);
923 /* unexpected pmd-mapped page? */
930 invalidation_needed
= true;
934 if (invalidation_needed
) {
935 mmu_notifier_invalidate_range(vma
->vm_mm
, address
,
936 address
+ (1UL << compound_order(page
)));
942 static bool invalid_mkclean_vma(struct vm_area_struct
*vma
, void *arg
)
944 if (vma
->vm_flags
& VM_SHARED
)
950 int page_mkclean(struct page
*page
)
953 struct address_space
*mapping
;
954 struct rmap_walk_control rwc
= {
955 .arg
= (void *)&cleaned
,
956 .rmap_one
= page_mkclean_one
,
957 .invalid_vma
= invalid_mkclean_vma
,
960 BUG_ON(!PageLocked(page
));
962 if (!page_mapped(page
))
965 mapping
= page_mapping(page
);
969 rmap_walk(page
, &rwc
);
973 EXPORT_SYMBOL_GPL(page_mkclean
);
976 * page_move_anon_rmap - move a page to our anon_vma
977 * @page: the page to move to our anon_vma
978 * @vma: the vma the page belongs to
980 * When a page belongs exclusively to one process after a COW event,
981 * that page can be moved into the anon_vma that belongs to just that
982 * process, so the rmap code will not search the parent or sibling
985 void page_move_anon_rmap(struct page
*page
, struct vm_area_struct
*vma
)
987 struct anon_vma
*anon_vma
= vma
->anon_vma
;
989 page
= compound_head(page
);
991 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
992 VM_BUG_ON_VMA(!anon_vma
, vma
);
994 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
996 * Ensure that anon_vma and the PAGE_MAPPING_ANON bit are written
997 * simultaneously, so a concurrent reader (eg page_referenced()'s
998 * PageAnon()) will not see one without the other.
1000 WRITE_ONCE(page
->mapping
, (struct address_space
*) anon_vma
);
1004 * __page_set_anon_rmap - set up new anonymous rmap
1005 * @page: Page to add to rmap
1006 * @vma: VM area to add page to.
1007 * @address: User virtual address of the mapping
1008 * @exclusive: the page is exclusively owned by the current process
1010 static void __page_set_anon_rmap(struct page
*page
,
1011 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1013 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1021 * If the page isn't exclusively mapped into this vma,
1022 * we must use the _oldest_ possible anon_vma for the
1026 anon_vma
= anon_vma
->root
;
1028 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1029 page
->mapping
= (struct address_space
*) anon_vma
;
1030 page
->index
= linear_page_index(vma
, address
);
1034 * __page_check_anon_rmap - sanity check anonymous rmap addition
1035 * @page: the page to add the mapping to
1036 * @vma: the vm area in which the mapping is added
1037 * @address: the user virtual address mapped
1039 static void __page_check_anon_rmap(struct page
*page
,
1040 struct vm_area_struct
*vma
, unsigned long address
)
1042 #ifdef CONFIG_DEBUG_VM
1044 * The page's anon-rmap details (mapping and index) are guaranteed to
1045 * be set up correctly at this point.
1047 * We have exclusion against page_add_anon_rmap because the caller
1048 * always holds the page locked, except if called from page_dup_rmap,
1049 * in which case the page is already known to be setup.
1051 * We have exclusion against page_add_new_anon_rmap because those pages
1052 * are initially only visible via the pagetables, and the pte is locked
1053 * over the call to page_add_new_anon_rmap.
1055 BUG_ON(page_anon_vma(page
)->root
!= vma
->anon_vma
->root
);
1056 BUG_ON(page_to_pgoff(page
) != linear_page_index(vma
, address
));
1061 * page_add_anon_rmap - add pte mapping to an anonymous page
1062 * @page: the page to add the mapping to
1063 * @vma: the vm area in which the mapping is added
1064 * @address: the user virtual address mapped
1065 * @compound: charge the page as compound or small page
1067 * The caller needs to hold the pte lock, and the page must be locked in
1068 * the anon_vma case: to serialize mapping,index checking after setting,
1069 * and to ensure that PageAnon is not being upgraded racily to PageKsm
1070 * (but PageKsm is never downgraded to PageAnon).
1072 void page_add_anon_rmap(struct page
*page
,
1073 struct vm_area_struct
*vma
, unsigned long address
, bool compound
)
1075 do_page_add_anon_rmap(page
, vma
, address
, compound
? RMAP_COMPOUND
: 0);
1079 * Special version of the above for do_swap_page, which often runs
1080 * into pages that are exclusively owned by the current process.
1081 * Everybody else should continue to use page_add_anon_rmap above.
1083 void do_page_add_anon_rmap(struct page
*page
,
1084 struct vm_area_struct
*vma
, unsigned long address
, int flags
)
1086 bool compound
= flags
& RMAP_COMPOUND
;
1091 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1092 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
1093 mapcount
= compound_mapcount_ptr(page
);
1094 first
= atomic_inc_and_test(mapcount
);
1096 first
= atomic_inc_and_test(&page
->_mapcount
);
1100 int nr
= compound
? hpage_nr_pages(page
) : 1;
1102 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1103 * these counters are not modified in interrupt context, and
1104 * pte lock(a spinlock) is held, which implies preemption
1108 __inc_node_page_state(page
, NR_ANON_THPS
);
1109 __mod_node_page_state(page_pgdat(page
), NR_ANON_MAPPED
, nr
);
1111 if (unlikely(PageKsm(page
)))
1114 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1116 /* address might be in next vma when migration races vma_adjust */
1118 __page_set_anon_rmap(page
, vma
, address
,
1119 flags
& RMAP_EXCLUSIVE
);
1121 __page_check_anon_rmap(page
, vma
, address
);
1125 * page_add_new_anon_rmap - add pte mapping to a new anonymous page
1126 * @page: the page to add the mapping to
1127 * @vma: the vm area in which the mapping is added
1128 * @address: the user virtual address mapped
1129 * @compound: charge the page as compound or small page
1131 * Same as page_add_anon_rmap but must only be called on *new* pages.
1132 * This means the inc-and-test can be bypassed.
1133 * Page does not have to be locked.
1135 void page_add_new_anon_rmap(struct page
*page
,
1136 struct vm_area_struct
*vma
, unsigned long address
, bool compound
)
1138 int nr
= compound
? hpage_nr_pages(page
) : 1;
1140 VM_BUG_ON_VMA(address
< vma
->vm_start
|| address
>= vma
->vm_end
, vma
);
1141 __SetPageSwapBacked(page
);
1143 VM_BUG_ON_PAGE(!PageTransHuge(page
), page
);
1144 /* increment count (starts at -1) */
1145 atomic_set(compound_mapcount_ptr(page
), 0);
1146 __inc_node_page_state(page
, NR_ANON_THPS
);
1148 /* Anon THP always mapped first with PMD */
1149 VM_BUG_ON_PAGE(PageTransCompound(page
), page
);
1150 /* increment count (starts at -1) */
1151 atomic_set(&page
->_mapcount
, 0);
1153 __mod_node_page_state(page_pgdat(page
), NR_ANON_MAPPED
, nr
);
1154 __page_set_anon_rmap(page
, vma
, address
, 1);
1158 * page_add_file_rmap - add pte mapping to a file page
1159 * @page: the page to add the mapping to
1161 * The caller needs to hold the pte lock.
1163 void page_add_file_rmap(struct page
*page
, bool compound
)
1167 VM_BUG_ON_PAGE(compound
&& !PageTransHuge(page
), page
);
1168 lock_page_memcg(page
);
1169 if (compound
&& PageTransHuge(page
)) {
1170 for (i
= 0, nr
= 0; i
< HPAGE_PMD_NR
; i
++) {
1171 if (atomic_inc_and_test(&page
[i
]._mapcount
))
1174 if (!atomic_inc_and_test(compound_mapcount_ptr(page
)))
1176 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
1177 __inc_node_page_state(page
, NR_SHMEM_PMDMAPPED
);
1179 if (PageTransCompound(page
) && page_mapping(page
)) {
1180 VM_WARN_ON_ONCE(!PageLocked(page
));
1182 SetPageDoubleMap(compound_head(page
));
1183 if (PageMlocked(page
))
1184 clear_page_mlock(compound_head(page
));
1186 if (!atomic_inc_and_test(&page
->_mapcount
))
1189 __mod_lruvec_page_state(page
, NR_FILE_MAPPED
, nr
);
1191 unlock_page_memcg(page
);
1194 static void page_remove_file_rmap(struct page
*page
, bool compound
)
1198 VM_BUG_ON_PAGE(compound
&& !PageHead(page
), page
);
1199 lock_page_memcg(page
);
1201 /* Hugepages are not counted in NR_FILE_MAPPED for now. */
1202 if (unlikely(PageHuge(page
))) {
1203 /* hugetlb pages are always mapped with pmds */
1204 atomic_dec(compound_mapcount_ptr(page
));
1208 /* page still mapped by someone else? */
1209 if (compound
&& PageTransHuge(page
)) {
1210 for (i
= 0, nr
= 0; i
< HPAGE_PMD_NR
; i
++) {
1211 if (atomic_add_negative(-1, &page
[i
]._mapcount
))
1214 if (!atomic_add_negative(-1, compound_mapcount_ptr(page
)))
1216 VM_BUG_ON_PAGE(!PageSwapBacked(page
), page
);
1217 __dec_node_page_state(page
, NR_SHMEM_PMDMAPPED
);
1219 if (!atomic_add_negative(-1, &page
->_mapcount
))
1224 * We use the irq-unsafe __{inc|mod}_lruvec_page_state because
1225 * these counters are not modified in interrupt context, and
1226 * pte lock(a spinlock) is held, which implies preemption disabled.
1228 __mod_lruvec_page_state(page
, NR_FILE_MAPPED
, -nr
);
1230 if (unlikely(PageMlocked(page
)))
1231 clear_page_mlock(page
);
1233 unlock_page_memcg(page
);
1236 static void page_remove_anon_compound_rmap(struct page
*page
)
1240 if (!atomic_add_negative(-1, compound_mapcount_ptr(page
)))
1243 /* Hugepages are not counted in NR_ANON_PAGES for now. */
1244 if (unlikely(PageHuge(page
)))
1247 if (!IS_ENABLED(CONFIG_TRANSPARENT_HUGEPAGE
))
1250 __dec_node_page_state(page
, NR_ANON_THPS
);
1252 if (TestClearPageDoubleMap(page
)) {
1254 * Subpages can be mapped with PTEs too. Check how many of
1255 * themi are still mapped.
1257 for (i
= 0, nr
= 0; i
< HPAGE_PMD_NR
; i
++) {
1258 if (atomic_add_negative(-1, &page
[i
]._mapcount
))
1265 if (unlikely(PageMlocked(page
)))
1266 clear_page_mlock(page
);
1269 __mod_node_page_state(page_pgdat(page
), NR_ANON_MAPPED
, -nr
);
1270 deferred_split_huge_page(page
);
1275 * page_remove_rmap - take down pte mapping from a page
1276 * @page: page to remove mapping from
1277 * @compound: uncharge the page as compound or small page
1279 * The caller needs to hold the pte lock.
1281 void page_remove_rmap(struct page
*page
, bool compound
)
1283 if (!PageAnon(page
))
1284 return page_remove_file_rmap(page
, compound
);
1287 return page_remove_anon_compound_rmap(page
);
1289 /* page still mapped by someone else? */
1290 if (!atomic_add_negative(-1, &page
->_mapcount
))
1294 * We use the irq-unsafe __{inc|mod}_zone_page_stat because
1295 * these counters are not modified in interrupt context, and
1296 * pte lock(a spinlock) is held, which implies preemption disabled.
1298 __dec_node_page_state(page
, NR_ANON_MAPPED
);
1300 if (unlikely(PageMlocked(page
)))
1301 clear_page_mlock(page
);
1303 if (PageTransCompound(page
))
1304 deferred_split_huge_page(compound_head(page
));
1307 * It would be tidy to reset the PageAnon mapping here,
1308 * but that might overwrite a racing page_add_anon_rmap
1309 * which increments mapcount after us but sets mapping
1310 * before us: so leave the reset to free_hot_cold_page,
1311 * and remember that it's only reliable while mapped.
1312 * Leaving it set also helps swapoff to reinstate ptes
1313 * faster for those pages still in swapcache.
1318 * @arg: enum ttu_flags will be passed to this argument
1320 static bool try_to_unmap_one(struct page
*page
, struct vm_area_struct
*vma
,
1321 unsigned long address
, void *arg
)
1323 struct mm_struct
*mm
= vma
->vm_mm
;
1324 struct page_vma_mapped_walk pvmw
= {
1330 struct page
*subpage
;
1331 bool ret
= true, invalidation_needed
= false;
1332 enum ttu_flags flags
= (enum ttu_flags
)arg
;
1334 /* munlock has nothing to gain from examining un-locked vmas */
1335 if ((flags
& TTU_MUNLOCK
) && !(vma
->vm_flags
& VM_LOCKED
))
1338 if (flags
& TTU_SPLIT_HUGE_PMD
) {
1339 split_huge_pmd_address(vma
, address
,
1340 flags
& TTU_MIGRATION
, page
);
1343 while (page_vma_mapped_walk(&pvmw
)) {
1345 * If the page is mlock()d, we cannot swap it out.
1346 * If it's recently referenced (perhaps page_referenced
1347 * skipped over this mm) then we should reactivate it.
1349 if (!(flags
& TTU_IGNORE_MLOCK
)) {
1350 if (vma
->vm_flags
& VM_LOCKED
) {
1351 /* PTE-mapped THP are never mlocked */
1352 if (!PageTransCompound(page
)) {
1354 * Holding pte lock, we do *not* need
1357 mlock_vma_page(page
);
1360 page_vma_mapped_walk_done(&pvmw
);
1363 if (flags
& TTU_MUNLOCK
)
1367 /* Unexpected PMD-mapped THP? */
1368 VM_BUG_ON_PAGE(!pvmw
.pte
, page
);
1370 subpage
= page
- page_to_pfn(page
) + pte_pfn(*pvmw
.pte
);
1372 if (!(flags
& TTU_IGNORE_ACCESS
)) {
1373 if (ptep_clear_flush_young_notify(vma
, pvmw
.address
,
1376 page_vma_mapped_walk_done(&pvmw
);
1381 /* Nuke the page table entry. */
1382 flush_cache_page(vma
, pvmw
.address
, pte_pfn(*pvmw
.pte
));
1383 if (should_defer_flush(mm
, flags
)) {
1385 * We clear the PTE but do not flush so potentially
1386 * a remote CPU could still be writing to the page.
1387 * If the entry was previously clean then the
1388 * architecture must guarantee that a clear->dirty
1389 * transition on a cached TLB entry is written through
1390 * and traps if the PTE is unmapped.
1392 pteval
= ptep_get_and_clear(mm
, pvmw
.address
,
1395 set_tlb_ubc_flush_pending(mm
, pte_dirty(pteval
));
1397 pteval
= ptep_clear_flush(vma
, pvmw
.address
, pvmw
.pte
);
1400 /* Move the dirty bit to the page. Now the pte is gone. */
1401 if (pte_dirty(pteval
))
1402 set_page_dirty(page
);
1404 /* Update high watermark before we lower rss */
1405 update_hiwater_rss(mm
);
1407 if (PageHWPoison(page
) && !(flags
& TTU_IGNORE_HWPOISON
)) {
1408 pteval
= swp_entry_to_pte(make_hwpoison_entry(subpage
));
1409 if (PageHuge(page
)) {
1410 int nr
= 1 << compound_order(page
);
1411 hugetlb_count_sub(nr
, mm
);
1412 set_huge_swap_pte_at(mm
, pvmw
.address
,
1414 vma_mmu_pagesize(vma
));
1416 dec_mm_counter(mm
, mm_counter(page
));
1417 set_pte_at(mm
, pvmw
.address
, pvmw
.pte
, pteval
);
1420 } else if (pte_unused(pteval
)) {
1422 * The guest indicated that the page content is of no
1423 * interest anymore. Simply discard the pte, vmscan
1424 * will take care of the rest.
1426 dec_mm_counter(mm
, mm_counter(page
));
1427 } else if (IS_ENABLED(CONFIG_MIGRATION
) &&
1428 (flags
& TTU_MIGRATION
)) {
1432 * Store the pfn of the page in a special migration
1433 * pte. do_swap_page() will wait until the migration
1434 * pte is removed and then restart fault handling.
1436 entry
= make_migration_entry(subpage
,
1438 swp_pte
= swp_entry_to_pte(entry
);
1439 if (pte_soft_dirty(pteval
))
1440 swp_pte
= pte_swp_mksoft_dirty(swp_pte
);
1441 set_pte_at(mm
, pvmw
.address
, pvmw
.pte
, swp_pte
);
1442 } else if (PageAnon(page
)) {
1443 swp_entry_t entry
= { .val
= page_private(subpage
) };
1446 * Store the swap location in the pte.
1447 * See handle_pte_fault() ...
1449 if (unlikely(PageSwapBacked(page
) != PageSwapCache(page
))) {
1452 page_vma_mapped_walk_done(&pvmw
);
1456 /* MADV_FREE page check */
1457 if (!PageSwapBacked(page
)) {
1458 if (!PageDirty(page
)) {
1459 dec_mm_counter(mm
, MM_ANONPAGES
);
1464 * If the page was redirtied, it cannot be
1465 * discarded. Remap the page to page table.
1467 set_pte_at(mm
, pvmw
.address
, pvmw
.pte
, pteval
);
1468 SetPageSwapBacked(page
);
1470 page_vma_mapped_walk_done(&pvmw
);
1474 if (swap_duplicate(entry
) < 0) {
1475 set_pte_at(mm
, pvmw
.address
, pvmw
.pte
, pteval
);
1477 page_vma_mapped_walk_done(&pvmw
);
1480 if (list_empty(&mm
->mmlist
)) {
1481 spin_lock(&mmlist_lock
);
1482 if (list_empty(&mm
->mmlist
))
1483 list_add(&mm
->mmlist
, &init_mm
.mmlist
);
1484 spin_unlock(&mmlist_lock
);
1486 dec_mm_counter(mm
, MM_ANONPAGES
);
1487 inc_mm_counter(mm
, MM_SWAPENTS
);
1488 swp_pte
= swp_entry_to_pte(entry
);
1489 if (pte_soft_dirty(pteval
))
1490 swp_pte
= pte_swp_mksoft_dirty(swp_pte
);
1491 set_pte_at(mm
, pvmw
.address
, pvmw
.pte
, swp_pte
);
1493 dec_mm_counter(mm
, mm_counter_file(page
));
1495 page_remove_rmap(subpage
, PageHuge(page
));
1497 invalidation_needed
= true;
1500 if (invalidation_needed
)
1501 mmu_notifier_invalidate_range(mm
, address
,
1502 address
+ (1UL << compound_order(page
)));
1506 bool is_vma_temporary_stack(struct vm_area_struct
*vma
)
1508 int maybe_stack
= vma
->vm_flags
& (VM_GROWSDOWN
| VM_GROWSUP
);
1513 if ((vma
->vm_flags
& VM_STACK_INCOMPLETE_SETUP
) ==
1514 VM_STACK_INCOMPLETE_SETUP
)
1520 static bool invalid_migration_vma(struct vm_area_struct
*vma
, void *arg
)
1522 return is_vma_temporary_stack(vma
);
1525 static int page_mapcount_is_zero(struct page
*page
)
1527 return !total_mapcount(page
);
1531 * try_to_unmap - try to remove all page table mappings to a page
1532 * @page: the page to get unmapped
1533 * @flags: action and flags
1535 * Tries to remove all the page table entries which are mapping this
1536 * page, used in the pageout path. Caller must hold the page lock.
1538 * If unmap is successful, return true. Otherwise, false.
1540 bool try_to_unmap(struct page
*page
, enum ttu_flags flags
)
1542 struct rmap_walk_control rwc
= {
1543 .rmap_one
= try_to_unmap_one
,
1544 .arg
= (void *)flags
,
1545 .done
= page_mapcount_is_zero
,
1546 .anon_lock
= page_lock_anon_vma_read
,
1550 * During exec, a temporary VMA is setup and later moved.
1551 * The VMA is moved under the anon_vma lock but not the
1552 * page tables leading to a race where migration cannot
1553 * find the migration ptes. Rather than increasing the
1554 * locking requirements of exec(), migration skips
1555 * temporary VMAs until after exec() completes.
1557 if ((flags
& TTU_MIGRATION
) && !PageKsm(page
) && PageAnon(page
))
1558 rwc
.invalid_vma
= invalid_migration_vma
;
1560 if (flags
& TTU_RMAP_LOCKED
)
1561 rmap_walk_locked(page
, &rwc
);
1563 rmap_walk(page
, &rwc
);
1565 return !page_mapcount(page
) ? true : false;
1568 static int page_not_mapped(struct page
*page
)
1570 return !page_mapped(page
);
1574 * try_to_munlock - try to munlock a page
1575 * @page: the page to be munlocked
1577 * Called from munlock code. Checks all of the VMAs mapping the page
1578 * to make sure nobody else has this page mlocked. The page will be
1579 * returned with PG_mlocked cleared if no other vmas have it mlocked.
1582 void try_to_munlock(struct page
*page
)
1584 struct rmap_walk_control rwc
= {
1585 .rmap_one
= try_to_unmap_one
,
1586 .arg
= (void *)TTU_MUNLOCK
,
1587 .done
= page_not_mapped
,
1588 .anon_lock
= page_lock_anon_vma_read
,
1592 VM_BUG_ON_PAGE(!PageLocked(page
) || PageLRU(page
), page
);
1593 VM_BUG_ON_PAGE(PageCompound(page
) && PageDoubleMap(page
), page
);
1595 rmap_walk(page
, &rwc
);
1598 void __put_anon_vma(struct anon_vma
*anon_vma
)
1600 struct anon_vma
*root
= anon_vma
->root
;
1602 anon_vma_free(anon_vma
);
1603 if (root
!= anon_vma
&& atomic_dec_and_test(&root
->refcount
))
1604 anon_vma_free(root
);
1607 static struct anon_vma
*rmap_walk_anon_lock(struct page
*page
,
1608 struct rmap_walk_control
*rwc
)
1610 struct anon_vma
*anon_vma
;
1613 return rwc
->anon_lock(page
);
1616 * Note: remove_migration_ptes() cannot use page_lock_anon_vma_read()
1617 * because that depends on page_mapped(); but not all its usages
1618 * are holding mmap_sem. Users without mmap_sem are required to
1619 * take a reference count to prevent the anon_vma disappearing
1621 anon_vma
= page_anon_vma(page
);
1625 anon_vma_lock_read(anon_vma
);
1630 * rmap_walk_anon - do something to anonymous page using the object-based
1632 * @page: the page to be handled
1633 * @rwc: control variable according to each walk type
1635 * Find all the mappings of a page using the mapping pointer and the vma chains
1636 * contained in the anon_vma struct it points to.
1638 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1639 * where the page was found will be held for write. So, we won't recheck
1640 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1643 static void rmap_walk_anon(struct page
*page
, struct rmap_walk_control
*rwc
,
1646 struct anon_vma
*anon_vma
;
1647 pgoff_t pgoff_start
, pgoff_end
;
1648 struct anon_vma_chain
*avc
;
1651 anon_vma
= page_anon_vma(page
);
1652 /* anon_vma disappear under us? */
1653 VM_BUG_ON_PAGE(!anon_vma
, page
);
1655 anon_vma
= rmap_walk_anon_lock(page
, rwc
);
1660 pgoff_start
= page_to_pgoff(page
);
1661 pgoff_end
= pgoff_start
+ hpage_nr_pages(page
) - 1;
1662 anon_vma_interval_tree_foreach(avc
, &anon_vma
->rb_root
,
1663 pgoff_start
, pgoff_end
) {
1664 struct vm_area_struct
*vma
= avc
->vma
;
1665 unsigned long address
= vma_address(page
, vma
);
1669 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1672 if (!rwc
->rmap_one(page
, vma
, address
, rwc
->arg
))
1674 if (rwc
->done
&& rwc
->done(page
))
1679 anon_vma_unlock_read(anon_vma
);
1683 * rmap_walk_file - do something to file page using the object-based rmap method
1684 * @page: the page to be handled
1685 * @rwc: control variable according to each walk type
1687 * Find all the mappings of a page using the mapping pointer and the vma chains
1688 * contained in the address_space struct it points to.
1690 * When called from try_to_munlock(), the mmap_sem of the mm containing the vma
1691 * where the page was found will be held for write. So, we won't recheck
1692 * vm_flags for that VMA. That should be OK, because that vma shouldn't be
1695 static void rmap_walk_file(struct page
*page
, struct rmap_walk_control
*rwc
,
1698 struct address_space
*mapping
= page_mapping(page
);
1699 pgoff_t pgoff_start
, pgoff_end
;
1700 struct vm_area_struct
*vma
;
1703 * The page lock not only makes sure that page->mapping cannot
1704 * suddenly be NULLified by truncation, it makes sure that the
1705 * structure at mapping cannot be freed and reused yet,
1706 * so we can safely take mapping->i_mmap_rwsem.
1708 VM_BUG_ON_PAGE(!PageLocked(page
), page
);
1713 pgoff_start
= page_to_pgoff(page
);
1714 pgoff_end
= pgoff_start
+ hpage_nr_pages(page
) - 1;
1716 i_mmap_lock_read(mapping
);
1717 vma_interval_tree_foreach(vma
, &mapping
->i_mmap
,
1718 pgoff_start
, pgoff_end
) {
1719 unsigned long address
= vma_address(page
, vma
);
1723 if (rwc
->invalid_vma
&& rwc
->invalid_vma(vma
, rwc
->arg
))
1726 if (!rwc
->rmap_one(page
, vma
, address
, rwc
->arg
))
1728 if (rwc
->done
&& rwc
->done(page
))
1734 i_mmap_unlock_read(mapping
);
1737 void rmap_walk(struct page
*page
, struct rmap_walk_control
*rwc
)
1739 if (unlikely(PageKsm(page
)))
1740 rmap_walk_ksm(page
, rwc
);
1741 else if (PageAnon(page
))
1742 rmap_walk_anon(page
, rwc
, false);
1744 rmap_walk_file(page
, rwc
, false);
1747 /* Like rmap_walk, but caller holds relevant rmap lock */
1748 void rmap_walk_locked(struct page
*page
, struct rmap_walk_control
*rwc
)
1750 /* no ksm support for now */
1751 VM_BUG_ON_PAGE(PageKsm(page
), page
);
1753 rmap_walk_anon(page
, rwc
, true);
1755 rmap_walk_file(page
, rwc
, true);
1758 #ifdef CONFIG_HUGETLB_PAGE
1760 * The following three functions are for anonymous (private mapped) hugepages.
1761 * Unlike common anonymous pages, anonymous hugepages have no accounting code
1762 * and no lru code, because we handle hugepages differently from common pages.
1764 static void __hugepage_set_anon_rmap(struct page
*page
,
1765 struct vm_area_struct
*vma
, unsigned long address
, int exclusive
)
1767 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1774 anon_vma
= anon_vma
->root
;
1776 anon_vma
= (void *) anon_vma
+ PAGE_MAPPING_ANON
;
1777 page
->mapping
= (struct address_space
*) anon_vma
;
1778 page
->index
= linear_page_index(vma
, address
);
1781 void hugepage_add_anon_rmap(struct page
*page
,
1782 struct vm_area_struct
*vma
, unsigned long address
)
1784 struct anon_vma
*anon_vma
= vma
->anon_vma
;
1787 BUG_ON(!PageLocked(page
));
1789 /* address might be in next vma when migration races vma_adjust */
1790 first
= atomic_inc_and_test(compound_mapcount_ptr(page
));
1792 __hugepage_set_anon_rmap(page
, vma
, address
, 0);
1795 void hugepage_add_new_anon_rmap(struct page
*page
,
1796 struct vm_area_struct
*vma
, unsigned long address
)
1798 BUG_ON(address
< vma
->vm_start
|| address
>= vma
->vm_end
);
1799 atomic_set(compound_mapcount_ptr(page
), 0);
1800 __hugepage_set_anon_rmap(page
, vma
, address
, 1);
1802 #endif /* CONFIG_HUGETLB_PAGE */